AI & Copyright Case Tracker

In an action brought by German music collecting society GEMA against Open AI, the Munich Regional Court ruled that

• ChatGPT stores reproductions of (copyright protected) song lyrics in its AI model; 
• it constitutes a “reproduction” if training data are contained in the model in a reproductible way, even if this requires further intermediate steps;
• although the outputs - which contained parts of the lyrics- were generated by end users' prompts, the defendant was seen as responsible due to its central role in creating the outputs.

Germany

Link

Link

In an action brought by Getty Images, an image library, against Stability AI, the developer of the Stable Diffusion text-to-image genAI model, the High Court found that:

• Getty Image's claim of trade mark infringement in respect of watermarked outputs was successful but only for earlier versions of the Stable Diffusion model – Getty had not shown that later versions of the model would produce infringing outputs in the real world;
• Getty Image's claim that Stability AI had committed an act of secondary copyright infringement (by importing an "article" into the UK that is an "infringing copy" of Getty's copyright works) failed - while an "article" can constitute an intangible object such as an AI model, where (on the claimant's own admission) that article does not - and never has - stored copies of the copyright works, it is not an "infringing copy"

Getty Images dropped its claim that Stability AI was liable for primary copyright infringement by copying Getty Image's copyright works during the training and development of Stable Diffusion – there was no evidence that relevant activities had occurred in the UK. Likewise, Getty Image's claim of output infringement was dropped, partly because Stability AI agreed to block certain prompts that led to allegedly infringing outputs and partly because it was not clear that any of those outputs copied a "substantial" part of any of Getty Image's copyright works.

UK

Germany

The plaintiff claims that standard terms and conditions in contracts between a newspaper and a freelance journalist

• that require the journalist to grant the newspaper rights to use journalists' work for AI training and use
• and permit the transfer of these rights of use without further remuneration should be seen as invalid unfair contract terms creating an unreasonable disadvantage for the journalist.

Germany

The Hungarian court requested a preliminary ruling by the European Court of Justice on the following questions:

• Communication to the public: Is it a copyright infringement if the chatbot shows parts of articles that are protected; if so, does it matter that the AI chatbot only predicts words based on patterns it has learned?
• Reproduction During Training: Does using patterns from existing texts during training qualify as reproduction, and is this allowed under the EU’s text and data mining exception?
• Reproduction by Output: Is it reproduction if the chatbot’s reply includes part or all of a press article based on a user's prompt?

Hungary

French collecting societies claim

• copyright infringement due to the alleged unauthorized use of copyright protected works for the purpose of training an AI model.
• economic unfairness (parasitism) due to the use of third party investment and creative efforts without contribution to their costs

France

German music collecting society GEMA claims

• that the AI tool Suno infringes copyright by generating content that closely resembles GEMA works
• that Suno used GEMA works for purposes of training its AI model without authorization.

Germany

• The downloading and storage of data in the search engine’s index for the purpose of creating AI generated snippets of articles is permitted under the TDM exception
• The plaintiff did not opt out of the TDM in the form required by law.

Hungary

Dutch media companies claimed that the AI created summaries of protected works infringed copyright. The court

• decided that the use was justified by legal exceptions, including the TDM exception
• stated that opt-out declarations must be specific and that an opt-out declared via a robots.txt protocol excluding only certain AI bots was not a valid opt-out from the TDM exception.

Netherlands

• The defendant’s use of the plaintiff’s photography (matching a downloaded image with pre-existing descriptions) qualified as text and data mining for purposes of scientific research, without a possibility to opt out
• The TDM exception likely also covers uses for the purpose of AI training
• An out-opt to TDM must be specific, but can be declared in natural language in the terms and conditions of a website

Germany

• Under Czech copyright law, only natural persons can be authors of copyright-protected works.
• An image generated by AI based on a common prompt does not qualify for copyright protection

Czech Republic